Evaluating statistical downscaling for daily maximum and minimum temperatures in Argentina

Empirical statistical downscaling (ESD) under the perfect prognosis approach was carried out to simulate daily maximum (Tx) and minimum temperatures (Tn) in the different climatic regions of Argentina. In this regard, three ESD techniques: analogs (AN), generalized linear models (GLM) and neural networks (NN) were evaluated considering multiple predictor sets with a variety of configurations driven by three different reanalysis. ESD models were cross-validated with folds of non-consecutive years (1979-2014) and then evaluated in a warmer set of years ( 2015-2018). The focus of the assessment of the ESD models was put on some marginal and temporal aspects of Tx and Tn. Depending on the aspect analyzed, AN ,GLM or NN models were more/less skillful but no method fulfilled all the features of both predicand variables. In this sense, the predictor set and model configuration were key factors. The different predictor structures (point-wise, spatial-wise and combinations of them) introduced the main differences for each ESD method, regardless of the predictand variable, region and reanalysis choice. In addition, the differences observed in ESD models due to the reanalysis choice were notably lower than the ones obtained due to changes in the statistical family and model structure. In the case of predictor variables, no improvements were observed in Tx and Tn simulations when a more complex predictor set was considered. In the case of Tn, models’ skills considerably increased when humidity information was included in the predictor set.  Our results showed that downscaling models were able to capture the general characteristics of Tx and Tn in all regions, with better performance in the latter variable. Overall, promising results were obtained in the evaluation of the ESD models in Argentina which encourage us to continue exploring their potential in different applications.